I’ve designed RF and microwave filters for more than 35 years, and I have watched the industry repeatedly underestimate the need for filters. Looking at the current 6G roadmaps, I see this pattern repeating. But this time, these challenges will make every wireless standard before them seem relatively simple in comparison.
When we designed filters for early cellular systems at 800 and 90 MHz, we achieved insertion losses of 3 or 4 dB, and cavity filters provided more than adequate rejection. Today, 5G requires insertion losses of less than 2 dB and rejection of more than 40 dB. But the 6G standards propose that operation will enter the upper reaches of the millimeter-wave region and beyond.
Conductor losses scale with the square root of frequency. A filter topology that delivers 1 dB insertion loss at 30 GHz might exhibit 3 dB loss at 100 GHz using the same materials. At these frequencies, surface roughness — which is negligible at microwave frequencies — becomes a dominant loss mechanism because the skin depth approaches the size of the copper grain boundaries.
We also face a geometric conflict. I have designed cavity filters with unloaded Q-factors above 10,000, but they are physically large. Those dimensions do not fit in smartphones or millimeter-wave base station arrays, where element spacing is strictly limited by wavelength. We also cannot rely on bulk acoustic wave (BAW) and surface acoustic wave (SAW) filters, which revolutionized handsets. Their operating frequencies are limited to around 15 GHz due to material constraints in aluminum nitride and lithium niobate.
The industry has proposed alternatives like substrate-integrated waveguides (SIW), silicon micromachined cavities, and metamaterials. SIW filters typically achieve Q-factors of 200 to 500 at millimeter-wave frequencies, resulting in insertion losses that break link budgets. Micromachined air cavities struggle with manufacturing yield. Dielectric resonators offer high Q-factors but suffer from batch-to-batch variation. As for metamaterials, I have seen very few implementations that achieve compact size and adequate Q-factors simultaneously.
Integration presents another hurdle. Co-packaging filters with amplifiers requires isolation above 50 dB and careful management of parasitics. I have debugged enough modules to know that wire-bond inductance and solder-bump capacitance will shift filter responses. A filter showing 1.5 dB insertion loss on a test fixture often degrades to 2.2 dB after packaging. Electromagnetic simulation helps, but it does not capture every variable. Physical prototyping is still essential, and that takes time.
From a filter design standpoint, fundamental design theory does not disappear at these frequencies because coupled-resonator synthesis and network realizations still describe the desired transfer function. The issue is that what engineers mean by “standard topologies” — such as lumped element, microstrip, stripline, hairpins, interdigital structures — run into a steep wall set by loss, parasitics, and manufacturing tolerance.
For example, at 300 GHz the free-space wavelength is about 1 mm, so feature sizes and coupling gaps become so small that ordinary fabrication variation becomes extraordinarily challenging and package/transition discontinuities stop being “parasitic” and start behaving like unintended resonators.
In addition, conductor loss and surface roughness become dominant because the skin depth is minimal at these frequencies, so the effective RF resistance is driven by metallization quality and roughness statistics. This means that unloaded Q is reduced for many planar structures, forcing a more aggressive trade-off between selectivity and insertion loss.
So are the filter topologies we’ve been using still viable at these high frequencies? The answer is that RF and microwave structures tend to look less like standard filters and more like precision electromagnetic structures, such as micromachined waveguide resonators, and tightly controlled integrated-waveguide concepts. Consequently, filtering often becomes more dependent on packaging, transitions, and beamforming. System partitioning is adjusted so that the hardware is not asked to deliver “brick-wall” selectivity at sub-THz frequencies, as designers are accustomed to at microwave and millimeter-wave frequencies.
This brings me to my main concern: the timeline. The goal for the early deployment of 6G has been set for around 2030. In my experience, transitioning a filter technology from a research prototype to production takes 5 to 7 years. So, in practical terms, I would expect pre-commercial trials and niche deployments to become visible between 2028 and 2029, with the first commercial networks appearing between 2029 and 2031, and meaningful device penetration beginning in 2032. If the industry’s business case is driven more by enterprise/industrial automation, sensing-integrated communications, or “AI-native” network operations than by consumer applications, that could accelerate deployments.
I am not suggesting that solving these challenges will be impossible. I’ve spent my entire career solving problems that initially seemed intractable. However, the industry needs to make hard decisions about investment and realistic performance assumptions now. If we wait, 6G may fail to deliver its promised capabilities on schedule.